inventory write-downs in the semiconductor industry no...in this paper, we study issues related to...
TRANSCRIPT
IIMB-WP N0. 524
WORKING PAPER NO: 524
Inventory Write-downs in the Semiconductor Industry
Hui Chen University of Colorado, Boulder
hui.chen@business,uzh.ch
Sundaresh Ramnath University of Miami
Srinivasan Rangan Associate Professor
Finance & Accounting Indian Institute of Management Bangalore Bannerghatta Road, Bangalore – 5600 76
Ph: 080-26993468 [email protected]
Steve Rock University of Colorado, Boulder
Year of Publication – September 2016
Electronic copy available at: http://ssrn.com/abstract=2836431
1
Inventory Write-downs in the Semiconductor Industry
Hui Chen*, Sundaresh Ramnath †, Srinivasan Rangan ‡, and Steve Rock*
* University of Colorado, Boulder † University of Miami ‡ Indian Institute of Management, Bangalore
April, 2015
Abstract: We study motives for and impacts of management discretion in inventory valuation. The semiconductor industry, with continual output price declines and rapid product obsolescence, provides an ideal setting to examine managers’ inventory write-down and production decisions. In this context, we develop a measure of ‘excess inventory’ and find that inventory write-downs are strongly correlated with this measure. We also find that inventory write-downs are timed strategically in periods of poor performance consistent with ‘big bath’ incentives. We construct a proxy for abnormal write-downs, and find that it is positively associated with subsequent operating performance, and negatively associated with future write-downs. Neither analysts, nor investors appear to fully appreciate the predictable implications of abnormal write-downs for subsequent operating performance.
_____________________________ We thank seminar participants at Indian School of Business, Ohio State University, Oklahoma State University, SunYat-Sen University, University of California at Davis, University of Colorado, University of Connecticut, and University of Oregon for helpful comments; Dov Fischer and Kristin Luttecke for research assistance, and the University of Colorado and the University of Miami for financial support.
Electronic copy available at: http://ssrn.com/abstract=2836431
2
1. Introduction
Under Generally Accepted Accounting Principles (GAAP), firms are required to
implement the lower-of-cost-or-market (LCM) rule and to write down the value of inventory if
its market value falls below historical cost.1 Such write-downs lead to lower reported earnings
for the same period. Firms must consider multiple, potentially subjective factors that are relevant
to the market value of inventories in making write-down decisions. Examples of such factors
include an estimate of the hypothetical replacement cost of the inventory, forecast demand for
the firm’s products, and expected costs of disposal.
In this paper, we study issues related to subjectivity in inventory valuation. First,
following prior research on asset write-downs, we hypothesize and test whether firms exploit
subjectivity in their inventory write-down decisions. We do so by correlating inventory write-
downs with several proxies for opportunism. In conducting these tests, we recognize that write-
downs also reflect consideration of economic factors and attempt to control for these factors.
Second, we argue that there are predictable consequences for subsequent operating performance
when firms’ inventory write-downs deviate from an expected level in a given period. More
specifically, we use an inventory write-down expectation model based on Bernard and Noel
(1991) to compute a proxy for ‘abnormal’ write-downs. We then test for the relation between
‘abnormal’ write-downs and subsequent operating performance and write-downs. Finally, we
examine whether market participants appear to appreciate the predictable reversing nature of
accrual-based earnings management through inventory. To this end, we test whether subsequent
1 Accounting Research Bulletin 43 (1953) defines ‘market’ as current replacement cost, but that amount extends over the following range: the maximum is expected selling price less costs of completion and disposal (also known as net realizable value), the minimum is net realizable value less the mark-up to sales price.
3
analyst forecast errors and risk-adjusted returns are correlated with ‘abnormal’ inventory write-
down.
We conduct our tests using a panel of firm years from the semiconductor manufacturing
industry. The semiconductor industry is characterized by rapid innovation with new generations
of more powerful semiconductors (chips) continually entering the market place. Existing chips
tend to be priced downward to be competitive with the performance level and pricing of new
chips. Consequently, the economic magnitude and frequency of inventory write-downs in this
industry are larger than most other industries.2 The second distinguishing characteristic of the
semiconductor industry is its pronounced cyclicality. Cyclicality causes earnings to be unusually
volatile which in turn increases incentives to manage earnings. Thus, the semiconductor industry
provides an excellent setting within which to examine inventory write-down decisions.
Our design and methodological approach consists of the following choices. Our sample
consists of an unbalanced panel of semiconductor firms for the period 1993-2007. We read all
the 10-Ks / annual reports for our sample firm years and obtain data on whether a write-down
was recorded and, if so, its dollar amount. We use a two-stage approach to compute our measure
of excess inventory. In the first stage, we use a pooled time-series and cross-sectional regression
to estimate ‘required’ inventory levels based on company-specific, as well as industry-wide
factors. Specifically, for each year, we use the preceding ten years of available information to
year t-1 relative to the sample year to estimate the industry-level coefficients of our inventory
prediction model. In the second stage, using the estimated model coefficients and the related
independent variables, we predict the ‘required’ inventory level for each firm-year. We deem the
difference between the predicted and actual inventory level as excess inventory that will
2 Of all the industries in the U.S., the semiconductor industry experienced the largest decline in its producer price index over the last two decades.
4
potentially be written off in year t. We then conduct our first set of tests via regressions of
inventory write-downs on our ‘excess’ inventory variable, various proxies of managerial
opportunism, a real production activity management proxy from Roychowdhury (2006), and
other economic determinants. This enables us to revisit the issue of whether inventory write-
downs appear to be managed, a finding not established by prior research.
We then compare our ‘excess’ inventory measure to the actual write-down and label the
difference the ‘abnormal’ write-down in year t. We argue that abnormally high (low) write-
downs in year t are likely to reverse, leading to improvements (declines) in future operating
performance, measured by changes in gross margin and return on assets. Additionally, if firms
delay taking write-downs, then abnormally low write-downs in year t are likely associated with
higher future write-downs. In addition, we examine whether abnormal write-downs are
associated with analyst forecast errors and future stock returns. If analysts appropriately
incorporate information regarding future implications of ‘abnormal’ write-down activity into
their forecasts, there should be no relation between ‘abnormal’ write-downs and subsequent
forecast errors. Further, in an efficient market, information related to write-downs should be
immediately incorporated into the price, thus resulting in no correlation between ‘abnormal’
write-downs and future stock returns.
Several important findings emerge from our research. First, we find that our estimates of
excess inventory are highly correlated with actual inventory write-downs, validating our model
of ‘required’ inventory levels. Second, we find evidence consistent with managers’ ‘bath-taking’
incentives impacting write-down decisions – firms tend to take larger write-downs when pre-
write-down performance is poor. Further, we find modest evidence of firms avoiding write-
downs to avoid reporting losses. Third, we find that both abnormal inventory write-downs are
5
significantly positively related to subsequent changes in operating performance. Fourth, we find
that abnormal write-downs in year t are negatively related to subsequent write-downs. We
interpret this as evidence of firms delaying write-downs in year t and being forced to take write-
downs in year t+1. Fifth and finally, we find that neither analysts, nor investors seem to
understand the implications of abnormal write-downs for future performance in that, current
abnormal write-downs are correlated with the subsequent analyst forecast errors and subsequent
stock returns. Our returns results obtain after controlling for standard risk factors (size, book-to-
market-ratio, and momentum) as well as three other accrual variables that have been shown to
predict future returns: working capital accruals, changes in inventories, and special items (Sloan
(1996); Thomas and Zhang (2002); Dechow and Ge (2006)).
We contribute to the existing literature in the following ways. First, we add to the limited
prior research on earnings management through inventory write-downs by providing initial
evidence that managers are more likely to take write-downs when performance is very poor and
to avoid reporting a loss.3 Francis, Hanna, and Vincent (1996) (FHV, henceforth) study
inventory write-downs along with other asset write-down decisions; but they do not find
evidence consistent with managerial incentives influencing inventory write-downs. FHV
conclude that inventory write-offs are not driven by incentives, arguing that market values are
readily available and that the LCM rule provides guidance which diminishes the role for
judgment. Our findings, however, are consistent with managerial incentives to strategically use
inventory write-downs, perhaps for the managers’ interests.4 The contrast between FHV’s
findings and ours can be potentially attributed to the differences in sample composition. Their
3 Prior studies examine discretion over write-downs in other asset types including Strong and Meyer (1987), Elliot and Shaw (1988), Zucca and Campbell (1992), Rees, Gill, and Gore (1996), Riedl (2004), Beatty and Weber (2006), and Ramanna and Watts (2012). 4 This finding is consistent with Baldenius and Reichelstein (2005), who analytically demonstrate that managers’ decisions on inventory management depend on their performance metrics.
6
analysis is based on a sample drawn from different industries. In contrast, we focus on the
semiconductor industry where the magnitude and frequency of inventory write-downs is higher.
In addition, we collect our write-down data from 10-Ks, whereas FHV focus on public newswire
announcements of write-downs.
Second, we also contribute to the earnings management literature by studying accrual-
based earnings management through inventories. We develop an ‘abnormal’ inventory write-
down proxy that manifests a reversal property – it is positively correlated with future
profitability. Interestingly, our measure of abnormal write-down is negatively related to
subsequent actual write-downs, consistent with firms delaying recording write-downs. Our study
complements and extends a rich accounting literature on earnings management in single accounts
and specific industries. Beaver (1996) notes that context-based research can address specific
accounting issues that are more relevant to some firms (or industries) than to others. Healy and
Wahlen (1999) also argue that evidence related to specific forms of earnings management can
provide more information to standard setters.5 By focusing on a single industry and single
account, we exploit the advantage afforded by more carefully modeling managerial discretion,
and consequently reduce noise in our regressions. A cost of these choices is generalizability of
the results.
Third, we provide new evidence that market participants do not correctly process the
information in abnormal write-downs for future profitability changes. Beginning with Sloan
(1996), several studies document a negative relation between accruals and subsequent stock
5 Many prior studies adopt this approach and focus on a single industry or single account for this purpose el (e.g., Moyer (1990); Petroni (1992); Beatty, Chamberlain, and Magliolo (1995); Collins, Shackelford, and Wahlen (1995); Beaver and Engel (1996), Key (1997); Miller and Skinner (1998); Ayers (1998); Schrand and Wong (2003); Jackson and Liu (2010)). Similar to Beaver and Engel (1996), who focus on loan loss reserves in banks, we investigate a sample of firms where the accrual choice is likely to be important in a sample of firms with relatively similar characteristics.
7
returns. Thomas and Zhang (2002) and Chan, Chan, Jegadeesh, and Lakonishok (2006)
document that inventory changes are the primary cause of the negative relation between accruals
and future abnormal returns. We find that, at least in the semiconductor industry, it is earnings
management through inventory that is misinterpreted by analysts and mispriced by the market.
Specifically, the evidence on forecast errors indicates that analysts do not fully incorporate
information conveyed in ‘abnormal’ write-downs into their forecasts. Also, controlling for other
risk factors and accrual/inventory-related anomalies, ‘abnormal’ write-downs are related to
subsequent market returns.
2. Semiconductor Industry
The semiconductor industry is among the largest manufacturing industries in the United
States. Since the 1960s, semiconductor devices have found applications in a diverse set of
products that range from digital watches, computers, and cellular phones to medical equipment,
automobiles, and weapon systems. The manufacture and sale of semiconductors is so crucial to
the U.S. economy that the Congress holds hearings on the industry to discuss its impact on
national policy. Scalise (2004), in his Congressional Testimony, summarizes the importance of
the semiconductor industry to the U.S. economy.6 He states that, “[s]emiconductors are, in
effect, the brains and nerve center for almost all electronic products today and are thus at the
heart of the entire IT sector, enabling everything from advanced computers to medical equipment
to weapons systems and contributing $75 billion annually to U.S. GDP, more than any other
single manufacturing technology.”
From its inception in 1958, when the first integrated circuit was developed, the
semiconductor industry has been cyclical, with alternating periods of rapid and slow growth.
6 George Scalise was the President of the U.S. Semiconductor Industry Association (SIA), at the time of this testimony.
8
Despite these periods of slow growth, during the years 1975-2000, the industry recorded an
impressive 16.1 percent compound annual growth rate (SIA (2004)). However, the industry
recorded its worst collapse in 2001, when, after achieving more than $200 billion in worldwide
sales in 2000, it witnessed a steep decline in sales of over thirty percent. A sharp decline in sales
of personal computers, cellular phones, and networking and communications equipment in 2001
negatively affected the demand for semiconductors.
As a response to the lowering market demand in 2001, the semiconductor industry
arguably took the biggest collective write-down decision in history. Semiconductor firms, along
with other technology firms, took sizable inventory write-offs (e.g., Conexant Systems took a
$149 million write-off and Xilinx took a $32 million write-off). In fact, the write-downs were so
large that they were considered to be overly aggressive by investors and the SEC “was closely
watching (Pender, 2001).”
The second striking feature of the industry is the high levels of research and development
activity and the consequent rapid rate of technological obsolescence. The famous prediction by
Gordon Moore, Intel Corp’s co-founder, in 1965, informally known as Moore’s law, has been
fairly accurate – the number of transistors on a chip tends to double every eighteen to twenty-
four months.7 New generations of faster and more powerful chips enter the market very
frequently, rendering the previous generation obsolete. This provides the second reason for the
high incidence of inventory build-ups and more frequent write-downs in this industry.
The third key aspect of the semiconductor industry is that chip selling prices fall
continually. Aizcorbe (2002) documents that, for the years 1993-1999, these declines are
primarily related to technological innovations, and to a lesser extent to lower manufacturing
7 For example, in 1971, the Intel 4004 microprocessor contained 2,300 transistors; in 1998, the Pentium II contained 7.5 million transistors.
9
costs. In addition, she finds that some of these price declines are related to the pricing strategy of
Intel Corporation, the dominant player in the semiconductor industry: Intel reduced its mark-ups
over costs during the 1990s. Because selling prices can fall below costs quite often,
semiconductor firms, more than firms in most other industries, are likely to take frequent
inventory write-downs.
To provide evidence on declining chip prices, we obtain Producer Price Index (PPI) data
for the semiconductor industry from the website of the Bureau of Labor Statistics, www.bls.gov.
PPI is a family of indexes that measures the average selling prices received by producers of
goods and services in the United States. The Bureau computes these indexes from data obtained
from over 25,000 establishments that provide approximately 100,000 price quotations per month.
In Figure 1, we report the monthly values of the PPI for the semiconductor industry for the years
1990-2008. Consistent with Aizcorbe (2002), semiconductor prices have declined considerably,
with the PPI falling from 153.4 points in the beginning of 1990 to 51.4 points at the end of 2008.
The downward trend is steady, with price declines occurring in 168 of the 228 months over the
sample period, or nearly seventy-four percent of the time. The mean January–to–January annual
percentage decline in the index is 5.5 percent with the largest decline in prices occurring in 2008
(20 percent). Importantly, in untabulated analysis, we find that, among all industries in the U.S.,
the semiconductor industry experienced the largest price declines over this period. Producer
price indices for other industries generally increase (e.g., for electronic capacitors the series is
more variable, but has increased by slightly over one percent over the same period).
Lastly, semiconductor firms are highly capital intensive – their operation requires
substantial investments in capital equipment. Because firms need to recover the high cost of
their equipment, they tend to produce at capacity even in periods of low demand. This tendency,
10
coupled with the high levels of technological obsolescence leads to build-up of inventories
during periods of low demand.
In sum, the semiconductor industry is characterized by cyclicality, rapid technological
change, continual price declines, and high capital costs. These industry characteristics result in
high risks of excess capacity and technological obsolescence and thus increase the likelihood of
inventory write-downs. Thus, this industry provides an excellent setting to study accruals
management via inventory write-downs.
3. Related Literature and Hypotheses Development
In this section, we link write-downs to managerial incentives, future profitability,
subsequent analyst forecast errors, and future stock returns.
3.1. Inventory Write-downs and Managerial Incentives
Prior research studies managerial motivations to opportunistically write-down asset
values. Most of these studies focus on discretion related to write-downs of long-lived assets.
Strong and Meyer (1987) find that write-downs are associated with senior management turnover
in the year of the write-down. Zucca and Campbell (1992) and Rees, Gill, Gore (1996) find that
write-downs of long-lived assets are more frequent in periods of “below-normal” earnings (a big
bath). FHV find that proxies for incentives are related to goodwill write-downs and restructuring
charges, but not to property, plant, and equipment write-downs. Riedl (2004) finds that long-
lived asset write-downs have greater (lower) association with big bath behavior in the period
after (before) the introduction of the standard on asset impairments (FAS 121). Beatty and
Weber (2006) and Ramanna and Watts (2012) find that goodwill write-downs are influenced by
several incentive variables such as the magnitude of slack in the net worth covenant, risk,
earnings response coefficients, the presence of an earnings-based bonus plan, CEO tenure,
11
exchange listing requirements, and measures of flexibility in estimating the fair value of
reporting units’ net assets.
FHV is the only study that examines motivations for inventory write-downs as a separate
category.8 They analyze 142 inventory write-downs that are publicly announced via press
releases or through newswires in the years 1988-1992 and find that three incentive related
variables – management turnover, the propensity to take a big bath, and income smoothing, are
unrelated to inventory write-downs. FHV argue that the ready availability of market values for
inventories and the guidance provided by Accounting Research Bulletin No. 43 reduces the
scope for judgment in valuing inventories. They conclude that the limited role of judgment
causes managerial incentives to be unrelated to inventory write-downs.
While we do not question FHV’s findings for their multi-industry sample, we believe that
the semiconductor industry provides a more powerful setting to detect the influence of
managerial incentives on write-downs. The primary reason for this is that inventory write-down
estimates in this industry are partly based on future uncertain conditions. Consequently, the
potential for judgment and discretion related to write-down amounts is magnified. For example,
Cypress Semiconductor Corporation in its 10-Q for fiscal quarter ended March 31, 2007 states
that,“[w]e record inventory write-downs as a result of our normal analysis of demand forecasts
and the aging profile of the inventory. We record charges to cost of revenues to write down the
carrying values of our inventories when their estimated market values are less than their
carrying values. The inventory write-downs reflect estimates of future market pricing relative to
the costs of production and inventory carrying values and projected timing of product sales.”
8 Allen, Larson, and Sloan (2013) also examine inventory write-downs. Their focus is on investigating if write-downs represent reversal of prior estimation errors. They find that, for a cross-section of firms from different industries, their proxy for estimation error is significantly and positively correlated with inventory write-downs, and negatively (but not significantly) correlated with subsequent inventory write-downs.
12
Similarly, Vitesse Semiconductor Corporation in its 10-K for the fiscal year ended September
30, 2012, states that, “[w]e evaluate inventories for excess quantities and obsolescence. Our
evaluation considers market and economic conditions; technology changes, new product
introductions, and changes in strategic business direction; and requires estimates that may
include elements that are uncertain.” In light of the potential discretion that semiconductor
firms’ managers have over the timing and magnitude of write-downs, we hypothesize (in
alternative form):
H1: Inventory write-downs in the semiconductor industry are influenced by managerial incentives.
3.2. Inventory Write-downs, Future Profits, and Subsequent Write-downs
Very little evidence on the relation between write-downs and subsequent profitability is
available from prior research. Further, available evidence pertains to long-lived assets. Elliot and
Shaw (1988) find that median earnings divided by market value of equity increases subsequent to
the year of the write-down. Zucca and Campbell (1992) find that mean quarterly return on assets
for firms that recorded relatively large write-downs are significantly lower than that of a control
sample in the three years following the write-down year. Rees, Gill and Gore (1996) find that
industry-adjusted return on assets are not significantly different from zero for firms that recorded
a single write-down in their sample period, but negative and significant for firms that recorded
multiple write-downs.
In contrast to write-downs of long-lived assets that have multi-year effects, inventory
write-downs are likely to affect only subsequent-year profits because inventory holding periods
are usually much shorter. Specifically, for our semiconductor industry sample, the mean
(median) inventory holding period is 114 days (92 days). Holding other things constant, by
13
lowering the cost of inventory, write-downs result are likely to result in lower cost of goods sold
and higher margins in the subsequent year when the inventory is sold.9
Recent research on accruals has shown that different accrual components have different
implications for future earnings. Xie (2001) presents evidence that the abnormal accrual
component is less persistent than the normal accrual component. Additionally, Allen, Larson,
and Sloan (2013) find that normal accruals and estimation errors have different reversing
patterns. Therefore, we decompose write-downs into predicted and abnormal write-downs, and
in light of the evidence in Xie (2001), to enhance the power of our tests, we correlate abnormal
write-downs with subsequent profit margins. Further, because abnormal write-downs represent
an estimation error in year t, we interpret low abnormal write-downs as evidence of delay in
writing down inventories. Consequently, we expect it to be negatively related to actual write-
downs in the next year. Thus, we make the following two predictions:
H2: Abnormal inventory write-downs are positively related to future profit margins.
H3: Abnormal inventory write-downs are negatively related to subsequent inventory write-downs.
3.3. Inventory Write-downs and subsequent Analyst Forecast Errors and Stock Returns
Prior research provides evidence that analysts ignore publicly-available information in
forming earnings forecasts (e.g., Bradshaw, Richardson, and Sloan (2001) and Ikenberry and
Ramnath (2002)). Our second hypothesis, H2, predicts a positive relation between abnormal
write-downs and subsequent profits. If analysts do not incorporate this pattern into their
forecasts of future earnings, then abnormal write-downs will be positively related to subsequent
forecast errors. However, if analysts efficiently process accounting information, there should be
9 If write-downs are associated with excess inventories, in competitive markets, firms might have to lower selling prices to dispose the excess inventories in the following year causing margins to decline. This selling price effect would offset some or all of the increase in profit associated with write-down reversals.
14
no discernible relation between ‘abnormal’ inventory write-downs and subsequent forecast
errors. Thus, our fourth hypothesis stated in alternate form is:
H4: Abnormal inventory write-downs are positively related to subsequent analyst forecast errors.
A few prior studies have examined whether write-down announcements are followed by
abnormal stock returns. Strong and Meyer (1987) find that for a sample of 78 write-down
announcements from 1981-1985, the mean sixty-day post write-down announcement cumulative
abnormal return (CAR) is +1.88 percent (t-statistic = 1.63). Elliott and Shaw (1988) examine
208 write-off firms from 1982-1985 and report a median cumulative industry-adjusted returns of
-3.1 percent for the six-month period after the write-off announcement. Bartov, Lindahl, and
Ricks (1998) examine 184 write-down announcements for the years 1984 and 1985. They
document mean (median) cumulative beta-adjusted excess return of -12 percent (-7.3 percent) in
the one year following the write-down announcement.
Beginning with Sloan (1996), several studies have documented a negative relation
between accruals and future stock returns and have interpreted this finding as consistent with a
market that over-estimates the persistence of accruals. Thomas and Zhang (2002) find that the
negative relation between accruals and future abnormal returns documented by Sloan (1996) is
due mainly to inventory changes. They conjecture that this effect is in turn is caused by
investors not anticipating the reversal of earnings management related to inventories. We build
on this conjecture and examine the possibility that investors fail to anticipate the reversal in
profits in year t+1 that is caused by abnormal write-downs in year t. This leads to our fifth
hypothesis (in alternative form):
H5: Abnormal inventory write-downs are positively related to subsequent one-year stock returns.
15
4. Empirical Model and Data Sources
To test our first hypothesis, we model write-downs as a function of managerial incentives
and control variables. Our proxies for managerial incentives to manage earnings via write-
downs include: the tendency to take “big baths” when profits are low, income smoothing,
avoiding losses, top management turnover, public issuance of equity or debt, and avoiding costs
of violating debt covenants. Our control variables in inventory write-down regressions include
excess inventory levels, past performance, industry demand, lagged write-downs, abnormal
production, and firm size. Thus, our model is:
ACTWDt = γ0 + γ1BOT10t + γ2TOP10t + γ3NEGSPECt + γ4MBZEROt +
γ5∆MGMTt +γ6OFFERt + γ7LTDDTAt + γ8PREDWDt +γ9FYRETt +
γ10NEGRETt + γ11NEGRETt × FYRETt + γ12BTMt + γ13∆BTMt +
γ14∆ROAt + γ15PCHCSt + γ16LAGWDt + γ17ABPRODt + γ18LABPRODt
+ γ19LNSALESt + εt (1)
where:
ACTWDt = actual amount of write-down divided by sales (COMPUSTAT data item REVT) in year t;
BOT10t = indicator variable equal to one if UE is in the bottom decile, zero
otherwise, where UE is defined as year t operating income after depreciation (COMPUSTAT data item OIADP), but prior to inventory write-downs less year t-1 operating income after depreciation, divided by year t-1 total assets (COMPUSTAT data item AT);
TOP10t = indicator variable equal to one if UE is in the top decile, zero otherwise,
where UE is defined above; NEGSPECt = indicator variable equal to one if the company has other income-decreasing
special items (COMPUSTAT data item SPI) in year t, and 0 otherwise. MBZEROt = indicator variable equal to one if income before extraordinary items
(COMPUSTAT data item IB) scaled by beginning of year market capitalization is greater than or equal to 0.00 and less than 0.01;
16
∆MGMTt = indicator variable equal to one if there is a change in any of the top three
executive positions (chairman of the board, chief executive officer, or president) in year t or year t-1, and 0 otherwise;
OFFERt = indicator variable equal to one if the firm has one or more offerings (debt or
equity) during year t, and zero otherwise; LTDDTAt = long-term debt (COMPUSTAT data item DLTT) divided by total assets at
the end of year t; PREDWDt = predicted amount of write-down in year t from Eq (2) in the appendix;
FYRETt = size-adjusted return compounded over the twelve months ending on the fiscal year end date of the year t;
NEGRETt = indicator variable equal to one if FYRETt is less than zero and zero
otherwise; BTMt = ratio of pre-write-down book value of equity (COMPUSTAT data item
CEQ) to the market value of equity (COMPUSTAT data item PRCC_F × COMPUSTAT data item CSHO) at the end of year t;
∆BTMt = change of BTM from year t-1 to t;
ΔROAt = year t-1 to year t change in pre-write-down operating income after depreciation (COMPUSTAT data item OIADP) divided by average total assets in year t (COMPUSTAT data item AT);
LAGWDt = write-down in year t-1 scales by sales (COMPUSTAT data item REVT) in
that year; PCHCSt = percentage change in computer shipments for the fiscal-year from the US
Bureau of Labor Statistics; ABPRODt = abnormal production, captured by the residual from the year-by-year
estimation of the production model, Eq. (3), described in the Appendix; LABPRODt = the lagged value of ABPRODt;
LNSALESt = log of sales (COMPUSTAT data item REVT) in year t;
We next describe the model variables in Eq. (1) in greater detail. FHV argue that the
propensity to take write-downs is likely related to the operating performance in the write-down
17
year. Specifically, if income is already low and managers are implicitly or explicitly rewarded
based on the level of earnings, managers have incentive to “take a bath” (accelerate write-down
recognition) to increase the probability of enjoying rewards in the future.10 Alternatively, if
semiconductor firms are more concerned about reporting a smooth earnings stream (a difficult
task in a cyclical industry), we expect a positive relation between write-downs and pre-write-
down earnings.11 We distinguish between bath-taking and smoothing explanations by including
two variables, BOT10 and TOP10, in the model. We use a random walk expectation to model
earnings and define unexpected earnings (UE) as the pre-write-down operating income after
depreciation (OIADP) in year t, less the reported OIADP in year t-1, divided by total assets at the
end of year t-1.12 BOT10 is an indicator variable equal to one if UE is in the bottom decile, zero
otherwise; and TOP10 is an indicator variable equal to one if UE is in the top decile, zero
otherwise. Deciles are based on the year-by-year cross-sectional distribution of UE. If firms
tend to take earnings baths via inventory write-downs, then BOT10 will be positively related to
write-downs. If smoothing is the dominant motivating factor, then the sign of the coefficient on
TOP10 should be positive and BOT10 should be negative.
As an additional proxy for big bath behavior, following FHV, we include an indicator
variable, NEGSPEC, which equals one if the company reports other income-decreasing special
10 The vast majority of our inventory write-downs are included in cost of goods sold; hence, compared to write-downs of other asset classes, concerns about whether managers are compensated based on income before or after non-recurring charges are less relevant. 11 In their two-period model, Kirschenheiter and Melumad (2002) show that managers tend to under-report to the maximum possible extent when earnings news is sufficiently bad and smooth earnings when earnings news is good. 12 We use pre-write-down earnings in the current year because the write-down decision is made prior to the write-down itself. We do not adjust the prior year earnings for any write-downs because managers are more likely to view reported earnings, as opposed to earnings adjusted for operating items such as write-downs, in the previous year, as the benchmark for the current year.
18
items (charges) in year t. We expect NEGSPEC to be positively related to inventory write-downs
consistent with write-downs being motivated as part of a broader “big bath” decision.13
Dechow and Skinner (2000), in their commentary on earnings management, argue that
managers have strong incentives to manage earnings by beating benchmarks. Survey evidence in
Graham, Harvey, and Rajagopal (2006) suggests that managers are willing to even forgo positive
NPV projects to beat benchmarks. We use the zero benchmark as our proxy and define
MBZERO as an indicator variable equal to one if income before extraordinary items scaled by
beginning-of-year market capitalization is in the range [0.00, 0.01). This approach is similar to
Roychowdhury’s (2006) identification of ‘suspect years’ to investigate real earnings
management.
We expect that management changes are likely to precipitate inventory write-downs. In
our setting, it seems reasonable that new management would benefit by writing down inventory
and attributing the charge to prior management. If the cost of the inventory sold during the new
management regime is artificially reduced based on a generous write-down, higher profits will be
reported when the chips are sold under current management, ceteris paribus. Consistent with
FHV, we include a dummy variable that equals one when there is a change in any of the top three
executive positions (chairman of the board, chief executive officer, or president) in the year of,
or the year prior to, the write-down (ΔMGMT). We expect ΔMGMT to be positively related to
inventory write-downs.
Firms that are considering public issues of debt or equity may have an incentive to inflate
earnings to obtain favorable prices on stock issuances or favorable credit terms on debt issues
(Teoh, Welch, and Wong, 1995). We obtain issuance data from the SDC database and define the
13 We note support for this argument in the popular press. Pender (2001) suggests that “[i]f a company announces a big inventory write-off along with restructuring charges, it can probably persuade analysts to disregard the inventory write-off, too.” We also use a continuous measure of special item charges and find similar results.
19
variable OFFER as 1 if the firm has one or more offerings (debt or equity) during year t, and
zero otherwise. We expect OFFER to be negatively related to inventory write-downs.
Defond and Jiambalvo (1994) present evidence that firms manage earnings via accruals
in response to their debt contracts. More recently, Beatty and Weber (2006) argue that firms that
are close to violating debt covenants are less likely to record goodwill impairment charges.
Therefore, we predict that firms time inventory write-downs so as to avoid violating debt
covenants. We measure proximity to debt covenant violation as long-term debt divided by total
assets (LTDDTA) in year t.14
Turning to our control variables, our first measure is the predicted write-down. Firms
record write-downs when there is a buildup of excess inventories and when market demand for
their products is expected to decline in the following year.15 We measure excess inventories
(PREDWD) as the difference between pre-write-down inventories and expected inventories. Our
measure of expected inventories is based on a time-series model for inventories developed by
Bernard and Noel (1992, BN henceforth). We expect PREDWD in year t to be positively related
to write-downs in that year.
To estimate excess inventory (PREDWD), we use a two-stage procedure. In the first
stage, for every year t, we use the previous ten years of data (year t-10 to t-1) for a sample of all
semiconductor firms (SIC 3674) and estimate a pooled, cross-sectional time-series regression of
inventory levels on its determinants. In the second stage, we compute the predicted inventory
level for each firm in year t as the product of the coefficients from the first stage regression and
the actual values of independent variables for each firm in year t. The difference between the
14 We considered using the debt-to-equity ratio, but do not do so because a non-trivial number of our firm-year observations have negative stockholders’ equity. 15 To better understand the determinants of write-downs we corresponded with a few semiconductor manufacturers. In general, our interviews suggest that write-down decisions are based on current inventory levels and assessments of expected market demand for firms’ products.
20
actual inventory level before the write-down and the predicted inventory is our estimate of
‘excess’ inventories or, alternatively, the predicted write-down. We repeat this procedure for
every year in our sample, thus allowing model coefficients in the first-stage regression to change
over time.
We employ the expectations model for inventory levels proposed by BN) to estimate the
first-stage estimation period regressions. BN use quarterly data in their study; we use their
independent variables but adapt their model to annual data because almost all write-down
disclosures are provided only in 10-Ks / annual reports. In the BN model, current inventory as a
percentage of sales is modeled as a function of its lagged value, current sales growth, lagged
sales growth, and lagged change in inventory divided by sales. In addition to the BN variables,
we include the annual change in the Producer Price Index (∆PPI) for the semiconductor industry
obtained from www.bls.gov. By doing so, we control for the impact of industry-level price
changes on firm-inventory levels.
Two clarifications are in order. First, we do not adjust estimation period inventories for
write-downs (i.e., we assume that firms do not use write-downs to manage earnings in the
estimation period). While this induces noise in inventory data for individual firm-years, we
expect that this noise is diversified away when we pool across firms and years. Second, we
focus our study on total inventories because inventory reserves and write-downs are typically not
delineated by type of inventory (i.e., raw materials, work-in-process, and finished goods).
Details of our adaptation of the BN model are provided in the Appendix.
We also include control variables related to a firm’s past stock market and accounting
performance to explain inventory write-downs. Stock prices are known to be leading indicators
of firm performance (Kothari and Sloan, 1992). Firms with poor stock price performance in the
21
current year are more likely to experience diminished sales and margins in the future. Such
firms are more likely to have impaired assets including overvalued inventory, and therefore are
more likely to write-down inventory in the current year. Accordingly, we expect stock returns
for the year t (FYRET) to be negatively related to the amount of the write-down. FYRET is
computed as the size-adjusted return, compounded over the twelve months ending with fiscal
year end date of year t.
Basu (1997) and others argue that firms use accrual adjustments to report conservatively;
that is, they are likely to record losses earlier than gains. To model this, consistent with Basu
(1997), we include (a) an indicator variable that equals one if FYRET is negative and zero
otherwise (NEGRET) and (b) the interaction between FYRET and NEGRET. If firms use write-
downs to report conservatively, we predict that the coefficient on the interaction term will be
negative.
A related measure of potentially overvalued assets is the firm’s book-to-market ratio.
Firms with high book-to-market ratios are more likely to have inventory with market prices
below cost. Therefore, we expect the ratio of book value of equity to market value of equity
(BTM) at the end of the year t to be positively related to inventory write-downs in year t. We add
back the tax-adjusted write-down amount to book value of equity to measure BTM prior to the
write-down decision. To capture the effect of relative declines in market value, we also include
the change in BTM (ΔBTM) from year t-1 to year t as an additional explanatory variable and
expect ΔBTM to be positively related to inventory write-downs.
Declining profitability is also an indicator of falling asset values; hence, we expect it to
increase the likelihood of inventory write-downs. We use a firm’s change in return on assets
(ΔROA) from year t-1 to year t as our measure of change in profitability and expect ΔROA to be
22
negatively related to inventory write-downs. ROA is defined as pre-write-down operating
income after depreciation divided by average total assets.
We control for industry-level demand in the year of the write-down. Because computer
manufacturers are the leading buyers of semiconductors, we include the year t percentage change
in computer shipments (PCHCS) obtained from the U.S. Bureau of Labor Statistics. Because a
positive PCHCS indicates increasing demand for computers, we predict that it is negatively
related to inventory write-downs.
To the extent that negative inventory and product impacts are persistent, write-downs are
likely to be positively serially correlated. Therefore, we include the lagged value of inventory
write-down as an independent variable (LAGWD) in the model.
Abnormal production (ABPROD) is likely to influence write-downs in two ways. First, if
production exceeds sales, a firm’s fixed costs will be spread over a larger number of units
leading to lower cost per unit. Because this is an accounting artifact, the market value per unit is
unlikely to be affected by this change in cost, per se. As market values have to “travel
downward” by a larger amount, increases in production reduce the likelihood of inventory write-
downs. Second, if firms deliberately increase production to lower cost of goods sold and thus
achieve their earnings targets, in general, the need to improve earnings through inventory write-
downs is reduced. Zang (2011) provides evidence consistent with this substitution effect; her
findings suggest that firms “fine-tune” accruals after the fiscal year-end based on the realized
real activities manipulation. On the other hand, producing more units could lead to excess
inventories and result in the need for subsequent write-downs, so the relation between abnormal
production and write-downs is equivocal. We also include lagged abnormal production costs in
our model (LAGABPROD) as excess production in the prior year could lead to subsequent write-
23
downs. We measure ABPROD and LAGABPROD following Roychowdhury (2006). Details of
this approach are provided in the Appendix.
Finally, we include the natural log of firm’s sales in year t (LNSALES) as an additional
factor that is potentially related to firms’ write-down decisions. Larger firms are likely to
possess products that are less prone to obsolescence and may also be able to withstand temporary
downturns in demand. We expect LNSALES to be negatively related to inventory write-downs.
We collect write-down amounts from 10-Ks in the SEC Filings Library of the LexisNexis
database and the Securities and Exchange Commission’s web site, www.sec.gov. We find write-
down data both from “Valuation and Qualifying Accounts,” alternatively reported on Schedule
II, VII, or VIII, typically at the end of the 10-K and by searching 10-Ks and annual reports for
the occurrence of the words “write” or “charge” or “adjust” with any suffix, within fifteen words
of “inventor”; the latter effectively captures potential variations of the word “inventory.”16
Where no write-down information is disclosed, we set the write-down amount to equal zero.
Our data sources for the independent variables in Eq. (1) are the University of Chicago’s
CRSP, COMPUSTAT, company 10-Ks, and the Bureaus of Labor Statistics’ web site. Data on
management changes are coded based on comparisons of top management lists of consecutive
annuals reports. We describe how we code write-downs in the next section.
5. Sample
The semiconductor industry can be broadly divided into three sectors: firms that are
primarily in the business of manufacturing and selling semiconductors, vertically-integrated
firms that produce semiconductors as inputs for products that they manufacture (e.g., IBM and
16 While write-downs are occasionally re-stated in subsequent years, our tabulated results are based on initially reported write-down amounts. We also perform analyses with restated amounts, and the results remain the same.
24
Sun Microsystems), and firms that do not manufacture semiconductors but which focus on
research and design of semiconductors (known as fabless firms, in industry parlance).
To arrive at our sample, we examine the CorpTech Directory of Technology Companies
for each of the years 1992-2007. This annual directory provides a comprehensive listing of all
firms in the semiconductor industry. Our initial sample consists of 220 firms that appeared in at
least one of the annual editions for the years 1992-2007, were domiciled in the U.S., were
publicly traded, and were not subsidiaries. From this initial sample, we exclude 73 vertically
integrated firms or firms that were not semiconductor manufacturers (based on our reading of
their 10-Ks), 13 fabless firms, 14 firms that are not listed on the CRSP database, and 8 firms for
which financial data was missing throughout our sample period on COMPUSTAT. We exclude
vertically integrated firms to achieve a sample of firms with fairly homogenous production
functions and fabless firms because they do not carry inventories. This yields a sample of 112
unique firms.
Our sample period consists of the 15 years, 1993-2007. The final sample consists of 861
firm-year observations for which we are able to determine the inventory write-down amounts
from firms’ form 10-Ks or annual reports and for which other requisite data are available on
CRSP and COMPUSTAT. The details regarding how we arrive at the final sample are presented
in panel A of table 1. The sample is an unbalanced panel, with the number of observations per
firm ranging from one to fifteen. The sample employed in our subsequent tests changes because
of additional data requirements; therefore, panel A also presents the sample sizes for all the
tables.
Panel B of table 1 provides descriptive statistics related to our sample. The mean
(median) firm-year level of sales in our sample is $979.72 ($208.68) million. The mean
25
(median) total assets is $1,478.04 ($310.03) million, and the mean (median) market
capitalization is $4,096.56 ($603.13) million. Reported end-of-year inventory across firm-years
has a mean (median) value of $105.77 ($27.52) million. The mean (median) write-downs scaled
by sales is 2.29% (0.00%) reflecting the fact that over 75% of the sample report no write-downs.
6. Results - Determinants of Write-downs
We model sales-deflated write-downs (ACTWD) as a function of managerial incentives
and other control variables. Before turning to the analyses of write-down determinants, we
present the coefficient estimates from the inventory model that we use to compute PREDWD
(Eq. (2)). Our model is a pooled, cross-sectional time-series regression of inventory levels on its
determinants. We estimate this model for every year t using the previous ten years of data (year
t-10 to t-1), with t ranging from 1993-2007. We also present the coefficients of the production
cost model Eq. (3) suggested by Roychowdhury (2006).
Panel A of Table 2 reports the average coefficients from the 10-year rolling regressions
and t-statistics from estimating Eq. (2) based on the distribution of the yearly coefficients (Fama
and MacBeth (1973)). The signs of the coefficients are consistent with results reported in BN
and all the coefficients are significant at less than the 1% level. Not surprisingly, the lagged
inventory-to-sales ratio is the most significant determinant of current year’s inventory-to-sales
ratio. Consistent with BN, we also find that current year’s inventory-to-sales ratio is negatively
related to current year’s sales growth, suggesting that inventory adjustment to sales is not
instantaneous. The average adjusted R2 from the regressions is 68.87%, indicating that the
model captures significant variation in inventory levels.
Panel B of Table 2 reports average coefficients from cross-sectional year-by-year
estimation of Eq. (3) following Roychowdhury (2006). We use residuals from this model as our
26
proxy for abnormal production costs. As with Roychowdhury, the most important factor in the
model by far is concurrent asset-scaled sales, and our average yearly coefficient for the
semiconductor industry (0.6346) is similar to the average in Roychowdhury estimated across
years and industries (0.7874).
We next analyze the determinants of inventory write-downs. In panel A of table 3, we
compare the mean and median values of independent variables in our write-down model across
the subsamples of firm-years with write-downs and those with no write-downs (345 and 348
observations, respectively). We use t-tests (Wilcoxon tests) to test for significant differences in
means (distributions).
The means and medians of ACTWD and PREDWD for the write-down group are both
significantly higher than those of the no-write-down group. ACTWD is the basis for defining the
two sub-samples and hence the result is obtained by construction. The difference between the
average amounts of PREDWD of the two sub-samples attests to the predictive ability of our
model of excess inventory. The means and medians of most of the other independent variables
differ between the write-down and the no-write-down subsamples largely in the expected
manner. The means and medians of both ABPROD and LAGABPROD are significantly higher
for the write-down group, which on the surface indicates that overproduction begets current and
subsequent write-downs. LAGWD values are higher for the write-down group. Both median and
mean FYRET (size-adjusted fiscal year returns) are significantly lower for the write-down sub-
sample than those of the no write-down sub-sample, supporting the hypothesis that firm stock
returns concurrently reflect or are leading indicators of write-downs. The proxy for conservatism
in reporting (NEGRET * FYRET) is more negative for the write-down subsample than the non-
write-down sample, which is indicative of conservatism in reporting. Consistent with
27
expectation, BTM (book-to-market ratio) and ∆BTM are significantly higher, and ∆ROA,
PCHCS, and LNSALES are lower for the write-down sub-sample.
In terms of incentive-related factors, both the mean and median LTDDTA are higher for
the write-downs group, which is inconsistent with an incentive to convey profitability when debt
is high. The difference in BOT10 between the two subsamples is marginally significant,
providing preliminary evidence that semiconductor firms are likely to take write-downs when
performance is already poor. The differences in TOP10 between the two samples are not
significant, implying that income smoothing is unlikely a primary incentive in our sample firms’
write-down decisions. The difference in NEGSPEC (an indicator variable signifying that a
negative special item was recognized during the year) suggests that firms that take charges in
either direction are more apt to have a concurrent negative special item charge, again confirming
the big-bath incentive. The average values of ∆MGMT, OFFER, MBZERO are not significantly
different between the two sub-samples. These univariate results are suggestive at best and the
multivariate analyses that we report next are intended to simultaneously control for non-strategic
(i.e., economic) and incentive-related motivations for write-downs.
We report results from the multivariate analysis of inventory write-downs in panel B of
table 3. We estimate model (1) using Tobit regressions as the dependent variable is non-negative
and has a preponderance of values that equal zero.17 Panel B of table 3 indicates that the
coefficients of PREDWD and LAGWD are both significantly positive, similar to the results from
the univariate analyses. The coefficient on PREDWD shows that our empirical estimate of the
write-down amount captures the potential overvaluation in pre-write-down inventory. ABPROD
is negative and significant, suggesting that overproduction and inventory write-downs are
17 We obtain similar results using OLS estimation.
28
substitute mechanisms to manage earnings. LAGABPROD is positively associated with inventory
write-downs, indicating that prior overproduction leads to subsequent write-downs..
Among the other economic factors, FYRET is significantly and negatively related to the
write-downs, confirming that stock performance is an important factor in inventory write-down
decisions. PCHCS is also negative and significant, indicating that a higher market demand is less
likely to lead to inventory write-downs. Similarly, LNSALES is also significant and negative, as
firms with higher sales are less likely to have the need to write down their excess inventory.
Among the incentive-related variables, BOT10 and NEGSPEC are both significantly
positive, providing evidence that our sample firms make write-down decisions consistent with
‘big bath’ motives. Since we find TOP10 is insignificant, we conclude that these sample firms
do not seem to use inventory write-downs for the purpose of income-smoothing. The results
indicate that MBZERO is negative and significant at the 10% level, suggesting that firms whose
scaled earnings ex-post marginally exceed zero are less likely to have taken inventory write-
downs. Evidence with respect to debt covenant violation avoidance is opposite to our
expectations in that LTDDTA is positively associated with write-down magnitudes. Controlling
for other factors, we do not find support for stock/debt issuance or management turnover driving
firms’ inventory write-down decisions.
Overall, our evidence on the determinants of write-downs suggests that firms are more
likely to take write-downs when they experience extremely poor performance (i.e., big bath
behavior). We also provide evidence that firms avoid write-downs to avoid reporting a loss.
7. Abnormal Inventory Write-downs, Subsequent Operating Performance and Subsequent Inventory Write-downs
We next examine how abnormal inventory write-downs relate to future operating
performance and future inventory write-downs. To conduct our tests, we require a proxy for
29
excess or abnormal write-downs. We measure abnormal write-down (ABWD) in any firm year
as the difference between the actual write-down in the year (ACTWD) and the predicted write-
down (PREDWD). The predicted write-down is our estimate of excess inventory computed as
the difference between the reported inventory-to-sales ratio and the predicted inventory-to-sales
ratio (see section 4 above).18 We expect ABWD in year t to be positively related to year t+1
change in both gross margin (COMPUSTAT data item GP) divided by sales (ΔGMt+1) and return
on assets (ΔROAt+1).
Panel A of table 4 contains univariate comparisons between firms that have positive
(HIGHWD) and negative (LOWWD) values of ABWD. Tests of significance are based on t-tests
for means and non-parametric Wilcoxon tests for distributions. The sample size is 668 firm
years, which is smaller than that of the previous table because we require subsequent operating
performance data. Of the 668 firm-year observations, a majority of the firms (389) have larger
than predicted write-downs, whereas 279 firm-year observations record smaller than predicted
write-downs. As expected, both mean and median year t+1 change in profitability, measured as
ΔGMt+1 or ΔROAt+1, are higher for the positive ABWD sub-sample. The univariate measures
support the hypothesis that year t abnormal write-downs reverse in the next year. Similarly, the
magnitude of inventory write-downs in the subsequent year (ACTWDt+1) is greater for LOWWD
firm-years than for HIGHWD firm-years, indicating that our proxy for abnormal write-downs is
negatively related to subsequent write-downs.
We also report univariate comparisons of other year t variables that are likely to influence
future performance: book-to-market ratio (BTMt), size-adjusted annual return (FYRETt), and
current change in profitability (ΔROAt and ΔGMt). As Panel A indicates, both mean and median
18 We do not use the data from table 3 to more finely separate write-downs into economic and opportunistic elements because we do not have a long time series of write-down determinants available for out of sample estimation.
30
BTM (ΔROAt) are lower (higher) for the HIGHWD sub-sample. Further, mean and median
FYRETt and ΔGMt are higher for the HIGHWD sub-sample.
To ensure that the effects of ABWD on subsequent year’s performance are incremental to
other factors, we include BTMt, FYRETt, and lagged change in performance as control variables
in multivariate regressions. In panel B of table 4 we present results from pooled regressions of
year t+1 operating performance measures on ABWD. Our dependent variables are change in
gross margin, change in ROA, and inventory write-downs (ACTWD), all measured in year t+1.
We expect that firms with lower (higher) than expected inventory write-downs in year t to report
lower (higher) profits and higher (lower) than expected write-downs in year t will report lower
(higher) write-downs in year t+1. Consistent with this expectation, we find that ABWD is
positively related to next year’s changes in gross margin and return on assets. Consistent with
prior research (Fairfield and Yohn, 2001), we also find that profitability changes in year t are
negatively related to profitability changes in year t+1, for both return on assets and gross margin.
We also find evidence that controlling for current write-downs, abnormal write-downs are
negatively-related to subsequent write-downs. This result is consistent with Allen et al. (2013)
who suggest that inventory write-down estimation error should be negatively related with
subsequent write-downs. Overall, the results in table 4 support the alternative hypothesis that
firms recording abnormally high or low write-downs experience profit reversals.19 The results
also indicate support for the alternative hypothesis that ‘over’- (‘under’) inventory write-downs
in one year manifests itself in lower (higher) write-downs in the next year.
8. Tests of market’s ability to process implications of write-downs for future performance
19 To confirm that there is no ‘mechanical effect’ of year t actual write-downs (ACTWD) impacting our subsequent performance results, we also run these models with ACTWD included as a regressor and our results are similar. The coefficients on ACTWD are uniformly not significantly different from zero.
31
We next examine whether market participants are able to understand the implications of
current period write-downs for future operating performance. We use analyst forecasts and stock
returns in the year following the write-down year to conduct two sets of tests.
8.1 Abnormal inventory write-downs and subsequent forecast errors
Prior research provides evidence that analysts ignore publicly-available information in
forming earnings forecasts (e.g., Bradshaw, Richardson, and Sloan, 2001 and Ikenberry and
Ramnath, 2002). In the previous section, we document a positive relation between abnormal
write-downs and subsequent operating performance. If analysts do not incorporate this pattern
into their forecasts of future earnings, then abnormal write-downs will be positively related to
subsequent forecast errors. However, if analysts efficiently process accounting information,
there should be no discernible relation between ‘abnormal’ inventory write-downs and
subsequent forecast errors.
To calculate analyst forecast errors for year t+1 earnings, we compute mean consensus
forecasts of annual earnings per share (EPS) for year t+1 from forecasts made within 60 days of
the beginning of month +4 relative to year end t. Our intent is to ensure that analysts have
information about year t inventory write-downs before making forecasts of year t+1 EPS. We
compute the difference between actual and forecast EPS from IBES and scale that difference by
price per share (also from IBES) to obtain our forecast error measure (AFEt+1). Prior research
indicates that forecast errors are serially correlated (Abarbanell and Bernard, 1992), so we
control for forecast errors of a comparable horizon from year t (AFEt).20 Prior research also
indicates that analysts are more optimistic from a temporal distance (e.g., Richardson, Teoh, and
Wysocki, 2004 and Lim, 2001). We control for forecast horizon (FLAGt+1) – the number of days
between the mean of the forecast dates comprising the consensus, and the earnings 20 We also run the forecast error model without lagged forecast errors and obtain consistent results.
32
announcement date per IBES. Additionally, we control for the number of forecasts comprising
the consensus (NFCSTt+1) and firm size (LNSALESt). Because we require two years of IBES
forecast data within a sixty-day interval from month t+4, and because not all sample firms are
covered in the IBES database, we have 472 firm year observations, fewer than the number of
observations in table 4.
We provide univariate and multivariate tests of analyst efficiency with respect to
information in abnormal write-downs. Consistent with our analysis of future operating
performance, we divide the sample into positive (HIGHWD) and negative (LOWWD) for
abnormal write-downs. Table 5, panel A indicates that year t+1 forecast errors for the LOWWD
sample are lower than those of the HIGHWD sample, consistent with the idea that analysts
under-react to information in abnormal write-downs. Results from both parametric and non-
parametric tests support this hypothesis.
In table 5, panel B, we regress year t+1 forecast errors (AFEt+1) on ABWDt, AFEt,
FLAGt+1, NFCSTt+1, and LNSALESt. We find that ABWDt is positively and significantly related
to subsequent forecast errors, after controlling for all other factors.21 The implication is that
analysts ignore information available in abnormal inventory write-downs which would improve
their forecasts. Alternately stated, analysts are surprised when abnormal write-downs in year t
result in earnings reversals in year t+1. Inconsistent with prior research, we note that forecast
errors appear negatively serially correlated in this sample. However, consistent with prior
research, forecast errors are significantly negatively related to FLAGt+1, consistent with analyst
optimism from a distance. Overall, our results support the alternative hypothesis that analysts
ignore information in abnormal write-downs.
21 We conduct rank (non-parametric) regressions and find very similar results.
33
8.2 Abnormal inventory write-downs costs and subsequent stock returns
In an efficient market, the implications of year t abnormal inventory write-down for
future profits will be priced immediately. Therefore, year t ABWD should be uncorrelated with
subsequent stock returns. However, if the market fails to appreciate the predictable performance
reversals of ABWD, subsequent returns will be positively correlated with lagged ABWD. That is,
firms that take excessive (deficient) write-downs in the current year will have higher (lower)
returns in the next year when they report predictably higher (lower) profits
Our future return variable is FUTRET, defined as the twelve month return from July of
year t to June of year t+1, compounded monthly. To ensure that information on ABWD is
available to investors before we begin cumulating returns, we measure it from the fiscal years
ending during the twelve-month period ending March of year t. Thus, the lag between fiscal year
end dates and the beginning of the return accumulation period ranges from three to fourteen
months. We are careful and follow the procedure suggested by Beavers, McNichols, and Price
(2007) to incorporate delisting returns when computing FUTRET.
To test H3, we estimate a pooled regression of FUTRET on lagged ABWD and control
variables. We include three controls for risk that are commonly employed in return prediction
regressions: logarithm of market capitalization measured at the end of June of year t (LOG
MCAP), the ratio of book equity value (after adding back the after-tax effect of the write-down)
to the market value of equity at the end of fiscal year in which ABWD is measured (BTM), and
the lagged value of FUTRET measured over the eleven month period ending at the end of May of
year t (LAGRET). Because change in pre-write-down ROA predicts subsequent profits, we
include it as an additional predictor. Beginning with Sloan (1996), several studies document that
accruals predict future returns. Thomas and Zhang (2002) show that a significant fraction of the
34
accrual effect is driven by the changes in inventories. Additionally, Dechow and Ge (2006)
show that special items have incremental explanatory power above accruals for future returns.
We therefore include lagged working capital accruals (WCACC), lagged change in inventories
(CHINV), and lagged special items (SPL_IT), each scaled by average assets as additional
controls.22
Table 6 contains our regression results. All standard errors are adjusted to account for
heteroscedasticity and clustering across firms at a point in time. The first column presents a
univariate regression of FUTRET on ABWD. Consistent with our expectation, ABWD is
positively and significantly related to FUTRET (coefficient = 1.30, t-statistic = 1.99). In the next
column, we regress FUTRET on ABWD and all our control variables. ABWD remains positive
and significantly related to future returns with a coefficient of 2.22 (t-statistic = 2.88). The
results indicate that ABWD have an economically important relation with FUTRET; a one
percent increase in abnormal write-downs (as a percentage of sales) is associated with a 2.2
percent increase in one-year-ahead returns.
Turning to the control variables, the coefficient on lagged market capitalization (LOG
MCAP) is negative and statistically significant (t-statistic = -4.16), indicating that large firms
perform worse than small firms. Consistent with the superior performance of ‘value’ stocks
documented in prior research (e.g., Lakonishok, Shleifer, and Vishny, 1994), lagged book-to-
market ratio is positively related to future returns (t-statistic = 1.76). Lagged stock returns are
negatively related to subsequent returns (t-statistic = -1.97), a result that is surprising in light of
the momentum effect documented by Jegadeesh and Titman (1993) and others. In contrast to
ABWD, neither working capital accruals nor inventory accruals are significantly related to 22 All three variables are obtained from COMPUSTAT. Working capital accruals are defined as (changes in current assets less change in cash) minus (changes in current liabilities – change in debt in current liabilities). Debt in current liabilities is set to zero when missing.
35
subsequent returns, using a 10% cut-off for statistical significance. Lastly, consistent with
Dechow and Ge (2006), SPL_IT has a negative relation with subsequent returns, but its
coefficient is not significant at the 10% level.
Overall, our results suggest that investors do not correctly understand the implications of
abnormal write-downs for future returns. Abnormal write-downs lead to predictable increases in
subsequent profits, and investors are surprised when these reversals occur.
9. Conclusion
We investigate the determinants and future consequences of inventory write-downs in the
semiconductor industry. The semiconductor industry is particularly interesting because of the
substantial fixed investment necessary to compete and corresponding tendency to produce at
capacity, its cyclical nature, and because unit values of inventory generally decrease over time,
making inventory write-downs more prevalent relative to most other industries. Overall, we
document that inventory levels in the semiconductor industry can be predicted using a simple
inventory model. We show that contemporaneous inventory write-downs are related to excess
inventories based on this model as well as incentives to take ‘big baths.’ Finally, we show that
abnormal write-downs are significantly correlated with future changes in profitability, write-
downs in subsequent years, analyst forecast errors, and risk-adjusted returns. Collectively, our
results provide new evidence that the stock market does not correctly process information in
abnormal write-downs for future profitability.
36
37
References
Abarbanell, J., and V. Bernard. 1992. Tests of analysts’ overreaction/underreaction to earnings
information as an explanation for anomalous stock price behavior. Journal of Finance,
47(3) 1181-1207.
Accounting Research Bulletin 43. 1953. Restatement and Revision of Accounting Research
Bulletins. Committee on Accounting Procedure.
Aizcorbe, A. 2002. Why are Semiconductor Prices Falling so Fast? Industry Estimates and
Implications for Productivity Measurement. FEDS Discussion Paper No. 2002-20.
Allen, E., C. Larson, and R. Sloan. 2013. Accrual reversals, earnings and stock returns. Journal
of Accounting and Economics 56(1) 113-129.
Ayers, B. 1998. Deferred tax accounting under SFAS No. 109: An empirical investigation of its
incremental value-relevance relative to APB No. 11. The Accounting Review 73 195-212.
Baldenius, T., and S. Reichelstein, 2005. Incentives for efficient inventory management: The role
of historical cost. Management Science 51(7) 1032-1045.
Bartov, E., F. Lindahl, and W. Ricks. 1998. Stock price behavior around announcements of
write-offs. Review of Accounting Studies 3(4) 327-346.
Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of
Accounting and Economics 24(1) 3-37.
Beatty, A., S. Chamberlain, and J. Magliolo. 1995. Managing financial reports of commercial
banks: The influence of taxes, regulatory capital and earnings. Journal of Accounting
Research 33 231-261.
Beatty, A., and J. Weber. 2006. Accounting discretion in fair value estimates: An examination of
SFAS 142 goodwill impairments. Journal of Accounting Research 44(2) 257-288.
38
Beaver, W. 1996. Directions in accounting research: NEAR and FAR. Accounting Horizons 10
113-124.
Beaver, W. and E. Engel. 1996. Discretionary behavior with respect to allowances for loan losses
and the behavior of security prices. Journal of Accounting and Economics 22(1) 177-206.
Beaver, W., M. McNichols, and R. Price. 2007. Delisting returns and their effect on accounting-
based market anomalies. Journal of Accounting and Economics 43(2) 341-368.
Bernard, V., and J. Noel. 1991. Do inventory disclosures predict sales and earnings? Journal of
Accounting Auditing and Finance 6(2) 145-181.
Bradshaw, M., S. Richardson and R. Sloan. 2001. Do analysts and auditors use information in
accruals? Journal of Accounting Research 39(21) 45-74.
Chan, K., L. Chan, N. Jegadeesh, and J. Lakonishok. 2006. Earnings quality and stock returns.
Journal of Business 79(3) 1041-1082.
Collins, J., D. Shackelford, and J. Wahlen. 1995. Bank differences in the coordination of
regulatory capital, earnings and taxes. Journal of Accounting Research 33 263-291.
Dechow, P. and W. Ge. 2006. The persistence of earnings and cash flows and the role of special
items: implications for the accrual anomaly. Review of Accounting Studies 11(2-3): 253-
296.
Dechow, P., and D. Skinner. 2000. Earnings management: reconciling the views of accounting
academics, practitioners, and regulators. Accounting Horizons 14(2) 235-250.
Defond, M. and J. Jiambalvo. 1993. Factors related to auditor‐client disagreements over income‐
increasing accounting methods. Contemporary Accounting Research 9(2) 415-431.
Elliot, J., and W. Shaw. 1988. Write-offs as accounting procedures to manage perceptions.
Journal of Accounting Research 26 91-119.
39
Fairfield, P. and T. Yohn. 2001. Using asset turnover and profit margin to forecast changes in
profitability. Review of Accounting Studies 6(4) 371-385.
Fama, E. and J. Macbeth. 1973. Risk, return, and equilibrium: empirical tests. Journal of
Political Economy 81 607-636.
Francis, J., J. Hanna, and L. Vincent. 1996. Causes and effects of discretionary asset write-offs.
Journal of Accounting Research 34 117-169.
Graham, J., C. Harvey, and S. Rajgopal. 2005. The economic implications of corporate financial
reporting. Journal of Accounting and Economics, 401() 3-73.
Healy, P. and J. Wahlen. 1999. Commentary: A review of the earnings management literature
and its implication for standard setting. Accounting Horizons 13(4) 365-383.
Ikenberry, D. and S. Ramnath. 2002. Underreaction to self-selected news events: The case of
stock splits. The Review of Financial Studies 15(2) 489-526.
Jackson, S., and X. Liu. 2010. The allowance for uncollectible accounts, conservatism, and
earnings management. Journal of Accounting Research 48(3) 565-601.
Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers: Implications
for stock market efficiency. The Journal of Finance 48(1) 65-91.
Key, K. 1997. Political cost incentives for earnings management in the cable television industry.
Journal of Accounting and Economics 23(3) 309-337.
Kirschenheiter, M., and N. Melumad. 2002. Can “big bath” and earnings smoothing co-exist as
equilibrium financial reporting strategies? Journal of Accounting Research 40(3) 761-
796.
Kothari, S., and R. Sloan. 1992. Information in prices about future earnings: Implications for
earnings response coefficients. Journal of Accounting and Economics 15(2) 143-171.
40
Lakonishok, J., A. Shleifer, and R. Vishny. 1994. Contrarian investment, extrapolation, and risk.
Journal of Finance 49(5) 1541-1578.
Lim, T. ‘Rationality and analysts’ forecast bias.’ Journal of Finance 56 (2001): 369-385.
McNichols, M., and G. Wilson. 1988. Evidence of earnings management from the provision for
bad debts. Journal of Accounting Research 26 1-31.
Miller, G. and D. Skinner. 1998. Determinants of the valuation allowances for deferred tax assets
under SFAS No. 109. The Accounting Review 73 213-233.
Moyer, S. 1990 Capital adequacy ratio regulations and accounting choices in commercial banks.
Journal of Accounting and Economics 13(2) 123-154.
Pender, K. ‘Write-offs remove excess inventory from books -- not shelves / Accounting move
can often distort firms’ financial data.’ The San Francisco Chronicle (May 8, 2001).
Petroni, K. 1992. Optimistic reporting in the property casualty insurance industry. Journal of
Accounting and Economics 15(4) 485-508.
Ramanna, K., and R. Watts. 2012. Evidence on the use of unverifiable estimates in required
goodwill impairment. Review of Accounting Studies 17(4) 749-780.
Rees, L., S. Gill, and R. Gore. 1996. An investigation of asset write-downs and concurrent
abnormal accruals. Journal of Accounting Research 34 157-170.
Richardson, S., S. Teoh and P. Wysocki. 2004. The walk-down to beatable analyst forecasts: The
role of equity issuance and insider trading incentives. Contemporary Accounting
Research 21(4) 885-924.
Riedl, E. 2004. An examination of long-lived asset impairments. The Accounting Review 79(3)
823-852.
41
Roychowdhury, S. 2006. Earnings management through real activities manipulation. Journal of
Accounting and Economics 42(3) 335-370.
Scalise, G. ‘Statement before the Committee on Financial Services, Subcommittee on Capital
Markets, Insurance and Government Sponsored Enterprises.’ U.S. Congress. Washington
D.C. (April 21, 2004).
Schrand, C., and F. Wong. 2003. Earnings management using the valuation allowance for
deferred tax assets under SFAS No. 109. Contemporary Accounting Research 20(3) 579-
611.
Semiconductor Industry Association (SIA). Annual Report. www.sia-online.org. San Jose, CA,
2004.
Sloan, R. 1996. Do stock prices fully reflect information in accruals and cash flows about future
earnings? The Accounting Review 71 289-315.
Strong, J. and J. Meyer. 1987. Asset writedowns: Managerial incentives and security returns.
Journal of Finance 42(3) 643-61.
Teoh, S., I. Welch, and T. Wong. 1998. Earnings management and the long‐run market
performance of initial public offerings. Journal of Finance 53(6) 1935-1974.
Thomas, J., and H. Zhang. 2002. Inventory changes and future returns. Review of Accounting
Studies 7(2-3) 163-187.
Xie, H. 2001. The mispricing of abnormal accruals. The Accounting Review 76(3) 357-373.
Zang, A. 2011. Evidence on the trade-off between real activities manipulation and accrual-based
earnings management. The Accounting Review 87(2) 675-703.
Zucca, L., and D. Campbell. 1992. A closer look at discretionary writedowns of impaired assets.
Accounting Horizons 6(3) 30-41.
42
Appendix 1
A. Model for Excess Inventories (PREDWD)
Our first-stage pooled cross-section time series model for inventory is:
𝐼𝐼𝑡𝑡𝑆𝑆𝑡𝑡
= α0 + α1 �1𝑆𝑆𝑡𝑡� + α2 𝐼𝐼𝑡𝑡−1
𝑆𝑆𝑡𝑡−1+ α3∆ �
𝐼𝐼𝑡𝑡−1𝑆𝑆𝑡𝑡−1
� + α4 𝑆𝑆𝑆𝑆𝑡𝑡 + α5 𝑆𝑆𝑆𝑆𝑡𝑡−1 + α6 ∆ 𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 + εt (2)
where firm subscripts are omitted for brevity, and:
It = reported inventory at the end of year t (COMPUSTAT data item INVT);
St = sales for year t (COMPUSTAT data item REVT);
∆ �𝐼𝐼𝑡𝑡−1𝑆𝑆𝑡𝑡−1
� = �𝐼𝐼𝑡𝑡−1𝑆𝑆𝑡𝑡−1
� − �𝐼𝐼𝑡𝑡−2𝑆𝑆𝑡𝑡−2
� is the change in scaled inventory from year t-2 to t-1;
SGt = 𝑆𝑆𝑡𝑡−𝑆𝑆𝑡𝑡−1 𝑆𝑆𝑡𝑡−1
is the percentage growth in sales from year t-1 to t;
SGt-1 = 𝑆𝑆𝑡𝑡−1−𝑆𝑆𝑡𝑡−2 𝑆𝑆𝑡𝑡−2
is the percentage growth in sales from year t-2 to t-1;
∆𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 = percentage change in the semiconductor industry producer price index from
year t-1 to t.
In the second stage, we collect the estimated coefficients from Eq. (2) and predict the
expected inventory level for year t. For example, to predict the inventory level 𝐼𝐼𝑡𝑡𝑆𝑆𝑡𝑡
for firm i in
1996 (t=1996), we use data for all available firm-years from 1986 to 1995 and estimate the
coefficients in equation (1). We then use the estimated coefficients and predict 𝐸𝐸 �𝐼𝐼1996𝑆𝑆1996
� for firm
i as follows:
𝐸𝐸 �𝑃𝑃1996𝑆𝑆1996
� = α0� + α1� �1
𝑆𝑆1996� + α2�
𝑃𝑃1995𝑆𝑆1995
+ α3�∆ �𝑃𝑃1995𝑆𝑆1995
� + α4� 𝑆𝑆𝑆𝑆1996 + α5� 𝑆𝑆𝑆𝑆1995
+ α6� ∆ 𝑃𝑃𝑃𝑃𝑃𝑃1996
43
We use the predicted value of inventory to sales 𝐸𝐸 �𝐼𝐼𝑡𝑡𝑆𝑆𝑡𝑡� described above and compute excess
inventory as 𝐼𝐼∗𝑡𝑡
𝑆𝑆𝑡𝑡 - 𝐸𝐸 �𝐼𝐼𝑡𝑡
𝑆𝑆𝑡𝑡� where 𝑃𝑃∗𝑡𝑡 is the reported inventory for year t plus the inventory write-
down for year t (i.e., the pre-write-down inventory level). If the inventory expectation model
provides reasonable estimates of ‘required’ inventory and firms’ write-down decisions are
related to excess inventory, then the empirically estimated excess inventory, or predicted write-
down (PREDWD = 𝐼𝐼∗𝑡𝑡
𝑆𝑆𝑡𝑡 -𝐸𝐸 �𝐼𝐼𝑡𝑡
𝑆𝑆𝑡𝑡�), will be positively related to the actual write-down amounts.
B. Model for Abnormal Production
We estimate the “abnormal” amount of production (ABPROD) for our sample firms
through a two-stage approach. First, we follow Roychowdhury (2006) to estimate the expected
level of production costs. Then we use the regression coefficients from the first-stage regression
to compute the abnormal production costs. The first-stage regression model is:
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡𝐴𝐴𝑡𝑡−1
= 𝛽𝛽0 + 𝛽𝛽1 �1
𝐴𝐴𝑡𝑡−1� + 𝛽𝛽2
𝑆𝑆𝑡𝑡𝐴𝐴𝑡𝑡−1
+ 𝛽𝛽3 �∆𝑆𝑆𝑡𝑡𝐴𝐴𝑡𝑡−1
� + 𝛽𝛽4 �∆𝑆𝑆𝑡𝑡−1𝐴𝐴𝑡𝑡−1
� + εt (3)
where firm subscript “i” is suppressed, and:
PRODi,t = Production costs in year t, equal to cost of goods sold in year t (COGSi,t) plus
the change inventory from t-1 to t (∆INVi,t), all scaled by assets in t-1 (Ai,t-1);
Si,t = Revenue in year t;
∆Si,t = Change in revenue from year t-1 to year t;
∆Si,t-1 = Change in revenue from year t-2 to year t-1.
In the second stage, we use the estimated coefficients from equation (2) to predict the expected
production cost. Then we use the predicted production cost scaled by firm assets 𝐸𝐸 �𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡𝐴𝐴𝑡𝑡−1
� to
44
compute abnormal production cost, �𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡∗
𝐴𝐴𝑡𝑡−1� − 𝐸𝐸 �𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡
𝐴𝐴𝑡𝑡−1�, where 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡∗ is the reported
production cost for year t.
45
Figure 1
US Monthly Semiconductor Producer Price Index (PPI) for the years 1990-2008
Monthly Producer Price Index (PPI) data for the semiconductor industry are from the website of the Bureau of Labor Statistics, www.bls.gov. Base is December 1998.
0
20
40
60
80
100
120
140
160
180
1990
1995
2000
2005
PPI - Semiconductors 1990-2008
46
Table 1
Sample Selection and Descriptive Statistics
Panel A: Sample Selection
Panel B: Sample Descriptive Statistics
Variable Mean Std. Deviation
25th percentile
Median 75th percentile
Inventory ($ millions) 105.77 294.61 10.20 27.52 77.40 Sales ($ millions) 979.72 3,149 79.25 208.68 576.85 Total Assets ($ millions) 1,478.04 4,300 108.21 310.03 1,026.34 Market Value ($ millions) 4,096.56 16,841 150.22 603.13 2,188.85 Inventory Write-down ($ millions) 7.78 41.69 0.00 0.00 3.40
Inventory Write-down as a % of Sales
2.29% 5.18% 0.00% 0.00% 1.91%
Our initial sample is obtained the CorpTech Directory of Technology Companies for the years 1992-2007. It includes publicly-traded U.S firms but excludes (a) subsidiaries, (b) vertically integrated firms, (c) non-semiconductor firms (d) fabless firms, (e) and firms not listed on the Center for Research in Security Prices (CRSP) database. Our sample period consists of the 15 years, 1993-2007. The final base sample consists of 861 firm-year observations for which we are unambiguously able to determine the inventory write-down amounts from firms’ form 10-Ks or annual reports and for which other requisite data are available on CRSP and COMPUSTAT. Data for the descriptive statistics are obtained from COMPUSTAT and 10-Ks.
Number of unique firms in sample 112 Maximum firm-years possible for the years 1993-2007 1,680 Less: Firm-years for which form 10-K is unavailable (143) Less: Firm-years with no clear write-down information or amount (173) Less: Firm-years without inventory data (10) Less: Firm-years with insufficient data on CRSP and COMPUSTAT (493) Base Sample 861 Less: Firm-years without lagged write-down data (for Table 3) (168) Table 3 Sample (firm-year observations) 693 Less: Firm-years without subsequent performance data (193) Table 4 Sample (firm-year observations) 668 Less: Firm-years without analyst forecast data (389) Table 5 Sample (firm-year observations) 472 Less: Firm-years without accruals or subsequent return data (49) Table 6 Sample (firm-year observations) 812
47
Table 2 Panel A: Ending Inventory Level Estimation
𝐼𝐼𝑡𝑡𝑆𝑆𝑡𝑡
= 𝑏𝑏0 + 𝑏𝑏1 �1𝑆𝑆𝑡𝑡� + 𝑏𝑏2
𝐼𝐼𝑡𝑡−1𝑆𝑆𝑡𝑡−1
+ 𝑏𝑏3 ∆ �𝐼𝐼𝑡𝑡−1𝑆𝑆𝑡𝑡−1
� + 𝑏𝑏4 𝑆𝑆𝑆𝑆𝑡𝑡 + 𝑏𝑏5 𝑆𝑆𝑆𝑆𝑡𝑡−1 + 𝑏𝑏6 ∆ 𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 + 𝑒𝑒𝑡𝑡 (2)
For every year from 1992 to 2007, regression (1) is estimated using firm-year observations from the previous ten years. Reported coefficients are the mean of the 16 yearly estimates. T-statistics are computed based on the distribution of the 16 yearly coefficients (Fama-MacBeth (1973)). The adjusted R2 is the mean of the 16 yearly adjusted R2s. The minimum (maximum) number of observations used in the estimation is 360 (931) firm-years and the mean (median) is 732 (696). *** indicates that the coefficient is significantly different from zero at the p=0.01 level or less.
Ii,t is reported inventory for firm i in year t; Si,t is sales revenue for firm i in year t; ∆ �𝐼𝐼𝑡𝑡−1
𝑆𝑆𝑡𝑡−1� = �𝐼𝐼𝑡𝑡−1
𝑆𝑆𝑡𝑡−1� − �𝐼𝐼𝑡𝑡−2
𝑆𝑆𝑡𝑡−2� is the change in scaled inventory from t-2 to t-1;
SGt = 𝑆𝑆𝑡𝑡−𝑆𝑆𝑡𝑡−1 𝑆𝑆𝑡𝑡−1
is the percentage growth in sales from t-1 to t;
SGt-1 = 𝑆𝑆𝑡𝑡−1−𝑆𝑆𝑡𝑡−2 𝑆𝑆𝑡𝑡−2
is the percentage growth in sales from t-2 to t-1; ∆ 𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 is the percentage change in the semiconductor industry producer price index over the 12-month period from the end of fiscal year t-1 to end of fiscal year t.
Expected Sign Mean Coefficient
t-statistic
𝑏𝑏0 ? 0.0286 *** 17.25
𝑏𝑏1 ? 0.1079 *** 13.24
𝑏𝑏2 + 0.8346 *** 271.60
𝑏𝑏3 ? -0.1147 *** -12.06
𝑏𝑏4 - -0.0355 *** -17.08
𝑏𝑏5 + 0.0124 *** 6.84
𝑏𝑏6 + 0.0701 *** 2.57
Mean Adjusted R2 68.87%
48
Table 2 Panel B: Production Model Estimation
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡𝐴𝐴𝑡𝑡−1
= 𝑏𝑏0 + 𝑏𝑏1 �1
𝐴𝐴𝑡𝑡−1� + 𝑏𝑏2
𝑆𝑆𝑡𝑡𝐴𝐴𝑡𝑡−1
+ 𝑏𝑏3 �∆𝑆𝑆𝑡𝑡𝐴𝐴𝑡𝑡−1
� + 𝑏𝑏4 �∆𝑆𝑆𝑡𝑡−1𝐴𝐴𝑡𝑡−1
� + 𝑒𝑒𝑡𝑡 (3)
Regression (2) is estimated by year from 1993 to 2007. Reported coefficients are the mean of the 16 yearly estimates. T-statistics are computed based on the distribution of the 15 yearly coefficients (Fama-MacBeth (1973)). The adjusted R2 is the mean of the 15 yearly adjusted R2s. The minimum (maximum) number of observations used in the estimation is 58 (100) firm-years and the mean (median) is 83 (85). *** indicates that the coefficient is significantly different from zero at the p=0.01 level or less.
Firm subscript i is suppressed above. PRODi,t is production costs in year t, equal to cost of goods sold in year t (COGSi,t) plus the change inventory from t-1 to t (∆INVi,t), all scaled by assets in t-1 (Ai,t-1); Si,t is revenue in year t; ∆Si,t is the change in revenue from year t-1 to year t; ∆Si,t-1 is the change in revenue from year t-2 to year t-1.
Expected Sign
Mean Coefficient
t-statistic
𝑏𝑏0 ? -0.0931 *** -5.225
𝑏𝑏1 ? 1.171 *** 5.615
𝑏𝑏2 + 0.6346 *** 32.442
𝑏𝑏3 ? -0.0294 -0.952
𝑏𝑏4 - -0.0545 -0.617
Mean Adjusted R2 81.63%
49
Table 3
Determinants of Inventory Write-downs
Panel A: Univariate Comparisons Write-down
firm-years (N=345) No Write-down firm-
years (N=348)
Mean Median Mean Median ACTWD (% of Revenues)
4.705 *** 1.928 *** 0.000 0.000
PREDWD (% of Revenues)
5.041 *** 2.159 *** -0.362 -0.678
ABPROD 0.009 *** 0.009 *** -0.033 -0.042 LAGABPROD 0.007 *** 0.007 *** -0.026 -0.035 LAGWD(%) 3.599 *** 1.058 *** 0.681 0.000 FYRET (%) 17.314 * -12.375 *** 33.950 -0.664 NEGRET 0.600 ** 1.000 ** 0.514 1.000 NEGRET*FYRET -0.237 *** -0.124 *** -0.177 -0.007 BTM 0.589 *** 0.484 *** 0.454 0.354 Δ BTM 0.066 ** 0.043 ** 0.008 0.000 Δ ROA 0.029 *** 0.042 *** 0.078 0.097 PCHCS -0.012 *** -0.009 *** 0.016 0.009 LNSALES 5.175 *** 5.203 *** 5.596 5.499 Δ MGMT 0.328 0.000 0.319 0.000 OFFER 0.072 0.000 0.063 0.000 LTDDTA 0.098 ** 0.015 *** 0.073 0.005 MBZERO 0.046 0.000 0.055 0.000 BOT10 0.122 * 0.000 * 0.080 0.000 TOP10 0.093 0.000 0.112 0.000 NEGSPEC 0.591 *** 1.000 *** 0.434 0.000
***, **, * indicates that the value for write-down observations is significantly different from the value for the no-write-down firm-year observations at the 0.01, 0.05, or 0.10 level, respectively. Significance levels are based on the t-test for difference in means and the Wilcoxon test for difference in medians.
50
Table 3, continued Panel B: Tobit Model of Factors Associated with Inventory Write-downs
Variable Predicted Sign
Equation (1) Dependent Variable = ACTWD
Intercept ? -0.0055 (-0.48) PREDWD + 0.4609 *** (15.98) ABPROD ? -0.0852 *** (-3.44) LAGABPROD ? 0.0708 *** (3.33) Other Economic Factors LAGWD + 0.4802 *** (9.70) FYRET - -0.0068 ** (-2.44) NEGRET ? -0.0030 (-0.41) NEGRET*FYRET - 0.0020 (0.13) BTM + -0.0015 (-0.18) ΔBTM + 0.0054 (0.58) ΔROA - 0.0278 (1.47) PCHCS - -0.0505
(-2.35) **
LNSALES ? -0.0060 *** (-3.88) Incentive Related Factors Δ MGMT + 0.0060 (1.21) OFFER - 0.0083 (0.89) LTDDTA - 0.0379 ** (2.08) MBZERO - -0.0185 * (-1.74) BOT10 + 0.0341 *** (4.31) TOP10 + -0.0054 (-0.63) NEGSPEC + 0.0241 ***
(4.95) Pseudo R2 59.22%
51
The sample consists of 693 firm-year observations. The t-statistics are reported below parameter estimates. ACTWD is the actual inventory write-down for year t scaled by sales for year t; PREDWD is the predicted write-down scaled by sales for year t obtained using the coefficient estimates from table 2; ABPROD is abnormal production, captured by the residual from Eq. (3), year-by-year estimation of the production model; LAGABPROD is the lagged value of ABPROD; FYRET is the size-adjusted return compounded over the twelve months ending on the fiscal year end date of year t; NEGRET is an indicator variable set equal to one if FYRET is less than zero and zero otherwise; BTM is the ratio of book equity value (after adding back the after-tax effect of the write-down or reversing abnormal production, as applicable) to the market value of equity at the end of year t; ΔBTM is the change in BTM from year t-1 (using reported book values) to year t; ΔROA is the change in ROA from year t-1 to year t where ROA is pre-write-down operating income after depreciation (OIADP), divided by average total assets; PCHCS is percentage change in computer shipments for the fiscal-year from the US Bureau of Labor Statistics; LNSALES is the natural log of sales for the year t; ΔMGMT is an indicator variable set equal to one if any of the three top officers of the company changed in the current or previous year, zero otherwise; OFFER is an indicator equal to one if the firm has one or more offerings (debt or equity) during year t, zero otherwise; LTDDTA is long-term debt divided by total assets at the end of year t; MBZERO is an indicator variable equal to one if earnings before extraordinary items scaled by beginning-of-year market capitalization is greater than or equal to zero, but less than 0.01. BOT10 is an indicator variable equal to one if UE is in the bottom decile, zero otherwise, where UE is defined as year t operating income after depreciation (prior to inventory write-downs or abnormal production, as applicable) less year t-1 operating income after depreciation, divided by year t-1 total assets; TOP10 is an indicator variable equal to one if UE is in the top decile, zero otherwise, where UE is defined as above; and NEGSPEC is an indicator variable set equal to one if the firm has negative special items in year t, zero otherwise. All non-indicator variables are winsorized at 1% and 99%. Data on ACTWD and ΔMGMT are from 10-Ks; data on FYRET is from CRSP; financial statement data are from COMPUSTAT; and data for the OFFER variable is from SDC. Equation (1) is estimated via Tobit. ***, **, and * indicate that the coefficient is significantly different from zero at the p=0.01, 0.05 and 0.10 level.
52
Table 4 Abnormal Write-downs, Subsequent Performance, and Subsequent Write-downs
Panel A: Univariate Comparisons HIGHWD
(ABWD > 0) (N=389)
LOWWD (ABWD < 0)
(N=279) Mean Median Mean Median
ABWDt (% of Revenues) 3.50*** 2.41*** -4.45 -2.71 ΔGMt+1 0.016*** 0.009*** -0.015 -0.008 ΔROAt+1 0.002*** 0.003** -0.022 -0.014 ACTWDt+1 (% of Revenues) 1.566*** 0.000*** 2.710 3.685 FYRETt (%) 24.35 -7.19 32.23 -10.814 BTMt 0.463** 0.381** 0.533 0.439 ΔGMt 0.006 0.007* -0.005 0.000 ΔROAt -0.001* 0.003** -0.016 -0.009 Panel B: Multivariate Analysis of Abnormal Write-downs and Future Performance
ΔGMt+1 ΔROAt+1 ACTWDt+1
Intercept -0.009 -0.024*** 0.012***
(-1.19) (-3.02) (3.48) ABWDt 0.288*** 0.195*** -0.249***
(4.33) (2.62) (-7.51)
FYRETt 0.005 0.013*** -0.001
(1.42) (3.20) (-0.30)
BTMt 0.021* 0.023* 0.004 (1.79) (1.76) (0.65)
ΔGMt -0.210*** - - (-5.57)
ΔROAt - -0.197*** - (-4.78)
ACTWDt - - 0.277***
(12.07)
Adjusted R2 7.05% 4.13% 21.22%
The sample consists of 668 firm-year observations. In panel A, p-values for differences in means are based on two-sample t-tests and differences in medians are based on two-sample Wilcoxon tests. In panel B, the t-statistics are reported below parameter estimates. All variable values are winsorized at 1% and 99%. ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. ABWDt is the “abnormal” write-down of inventory in year t, defined as the actual write-down in year t less the predicted write-down (see table 3 for the definition of predicted write-down); ΔGMt+1 is the change in gross margin (as a percentage of sales) from year t to year t+1; ΔROAt+1 is the change in ROA from year t to year t+1 where ROA is pre-write-down operating income after depreciation (OIADP) divided by average total assets; ACTWDt+1 is inventory write-down in year t+1; FYRET is the size-adjusted return compounded over the twelve months ending on the fiscal year end date of year t; BTM is the ratio of book equity value (after adding back the after-tax effect of the write-down) to the market value of equity at the end of year t. Data on FYRET is from CRSP; and financial statement data are from COMPUSTAT.
53
54
Table 5 Abnormal Write-downs and Subsequent Forecast Errors
Panel A: Univariate Comparisons HIGHWD
(ABWD > 0) (N=272)
LOWWD (ABWD < 0)
(N=200)
Mean Median Mean Median ABWDt (% of Revenues) 2.70*** 2.08*** -3.67 -1.94 AFEt+1 (% of Price) 0.00*** 0.03*** -0.97 -0.24 AFEt (% of Price) -0.23 0.05*** -0.65 -0.27 FLAGt+1 280.4 277 280.7 277.5 NFCSTt+1 13.74 11 13.00 11 LNSALESt 6.05 5.84 6.04 5.94 Panel B: Regression of Forecast Errors on Past Abnormal Write-downs
Predicted Sign
Intercept 0.0454**
(2.01) ABWDt + 0.0649**
(2.32) AFEt + -0.0936**
(-1.97) FLAGt+1 - -0.0001**
(-1.97) NFCSTt+1 ? 0.0002
(0.84)
LNSALESt ? -0.0019 (-1.37)
Adjusted R2 1.92% The sample consists of 472 firm-year observations. In panel A, p-values for difference in means are based on two-sample t-tests and those for difference in medians are based on two-sample Wilcoxon tests. In panel B, the t-statistics are reported below parameter estimates. ***, ** indicate statistical significance at the one percent and five percent levels, respectively. ABWDt is the “abnormal” write-down of inventory in year t, defined as the actual write-down in year t less the predicted write-down; HIGHWD (LOWWD) are firm-years for which the write-down was higher (lower) than the predicted values in year t; AFEt+1 is the forecast error, the difference between the mean analyst forecast of annual earnings per share for year t+1 and the actual earnings per share (EPS) for that year, deflated by stock price; forecasts are made within 60 days following the end of the fourth month in fiscal year t+1. AFEt is the forecast error for the year t horizon constructed in the same manner; FLAGt+1 is the mean forecast lag (calendar days between the mean forecast date and earnings announcement date) of forecasts comprising the consensus for year t+1; NFCSTt+1 is the number of forecasts used to construct the mean forecast for year t+1 earnings; and LNSALESt is the natural log of sales in year t. All variable values are winsorized at one percent and ninety-nine percent. Data on write-downs are from 10-Ks; data on LNSALES is from COMPUSTAT; data on analyst forecasts, actual EPS, and stock price are from IBES.
55
Table 6
Abnormal Write-downs and Subsequent Stock Returns Dependent Variable: Annual Return compounded from July of year t to June of year t+1 Predicted Sign Intercept 0.273***
(8.67) 1.437***
(4.16) ABWD + 1.323*
(1.95) 1.937***
(3.23) LOG MCAP - -0.098***
(-4.04) BTM - 0.198*
(1.74) LAGRET + -0.069*
(-2.19) ∆ ROA + 0.643
(1.54) WCACC + -0.555
(-0.58) CHINV + 0.962
(0.70) SPL_IT + -0.674
(-1.22) Adjusted R2 0.30%
3.44%
Sample size 812 812 The dependent variable is the annual compounded stock return from July of year t to June of year t+1. ABWD is the “abnormal” write-down of inventory in year t, defined as the actual write-down in year t less the predicted write-down; LOG MCAP is the logarithm of the market capitalization measured on June 30th of year t; BTM is the ratio of book equity value (after adding back the after-tax effect of the write-down) to the market value of equity at the end of year t; LAGRET is the annual compounded stock return from June of year t-1 to May of year t; ΔROA is the change in ROA from year t to year t+1 where ROA is pre-write-down operating income after depreciation (OIADP) divided by average total assets. WCACC equals (Change in Current Assets – Change in Cash) – (Change in Current Liabilities – Change in Short-term Debt) for year t divided by average assets in year t. CHINV is the change in inventories in year t divided by average assets in year t. SPL_IT is special items in year t divided by average assets in year t. All the financial statement variables are from before March 31st of year t. All variable values are winsorized at one percent and ninety-nine percent. Data for financial statement variables are from COMPUSTAT or Form 10-Ks and stock return data are from CRSP. Standard errors account for heteroscedasticity and clustering across firms. ***, **, and * indicate that the coefficient is significantly different from zero at the p=0.01, 0.05 and 0.10 level.